Note: This report focuses on product definitions, classification provisions, special control requirements, and the implications for companies’ regulatory registration work.
Common Regulatory Conclusions
All four documents are final classification orders (Final Amendment; Final Order) issued by FDA for four specific medical device types. Through De Novo classification, each device type is determined to be Class II (special controls), and the corresponding classification requirements are formally codified into the Code of Federal Regulations (CFR). FDA concludes that general controls, together with the special controls defined in these orders, provide a reasonable assurance of safety and effectiveness for these devices, while reducing the regulatory burden that previously resulted from their automatic placement into Class III.
All four device types are classified as Class II (special controls).
All four documents state that FDA has not exempted these device types from 510(k) under Section 510(m) of the FD&C Act.
Therefore, all four device types remain subject to 510(k) premarket notification requirements; Class II status does not mean automatic exemption from 510(k).
The new classification provisions are added to 21 CFR Parts 878, 892, and 868.
| Document Number | Device Type | New Regulation | Classification | Effective Date |
|---|---|---|---|---|
| 2026-22179 | Software to assist in the diagnosis of suspected skin cancer lesions | 21 CFR 878.1830 | Class II | 2024-01-12 |
| 2026-22180 | Software to assist in referral of pulmonary fibrosis-related imaging findings | 21 CFR 892.2085 | Class II | 2024-01-12 |
| 2026-22181 | Spinal imaging system for neuraxial anesthesia procedures | 21 CFR 868.1985 | Class II | 2024-01-19 |
| 2026-22182 | Over-the-counter sleep apnea risk assessment device | 21 CFR 868.2378 | Class II | 2024-02-06 |
Note on dates: the dates in the table are FDA’s original De Novo classification effective dates, not dates on which classification began in 2026. All four final orders state that they become effective upon publication in the Federal Register.
2. What the Four Documents Say
2.1 Document 2026-22179: Software to Assist in the Diagnosis of Suspected Skin Cancer Lesions
Device definition: a prescription-use, software-assisted diagnostic device that uses algorithms to analyze optical or other physical characteristics of skin lesions and outputs a classification of the skin lesion. Its intended use is as an adjunctive second-read tool for physicians who have not been trained in the clinical diagnosis and management of skin cancer, after they have identified a suspicious skin lesion. The device is not intended to be used as a stand-alone diagnostic device and is not intended to confirm a clinical diagnosis.
| Control Item | FDA Special Control Requirement |
|---|---|
| Clinical performance | Demonstrate the expected clinical performance of the device in the intended patient population through premarket clinical performance validation and postmarket surveillance; FDA may determine whether postmarket surveillance data are needed based on the totality of premarket review information. |
| Clinical comparison | Demonstrate that users assisted by the device achieve diagnostic characterization accuracy for lesions within the indicated use that is superior to users without the device. |
| Patient coverage | Clinical testing should cover different skin phototypes, risk factors, and anatomical sites. |
| Sensitivity and specificity | For lesions with high metastatic potential, device output sensitivity must be at least 90%; if lower sensitivity is proposed, alternative clinical justifications must be provided. Clinical justification must be provided for specificity. |
| Nonclinical performance validation | Demonstrate compatibility between the device software and specific signal or image acquisition hardware, and describe the compatible hardware, procedures, pre-specified test protocols, and data sets. |
| Precision validation | Demonstrate repeatability and reproducibility of device performance across different operators and under challenging use conditions. |
| Safety and software | Perform electrical, mechanical, thermal safety, and electromagnetic compatibility testing; complete software verification, validation, and hazard analysis. |
| Biocompatibility and reprocessing | Patient-contacting components must demonstrate biocompatibility; reusable components must have validated reprocessing instructions. |
| Human factors | Demonstrate that intended users can use the device safely and correctly. |
| Labeling | Summarize standalone performance and clinical performance testing, sensitivity/specificity and confidence intervals, performance in relevant subgroups, the population used to train the algorithm, device limitations, and risks associated with interpretation of output, and state that the device must not be used as a stand-alone diagnostic device; provide maintenance and reprocessing instructions. |
Key point: this document explicitly requires that for lesions with high metastatic potential, device output sensitivity must reach at least 90%, or alternative clinical justifications supporting lower sensitivity must be provided.
2.2 Document 2026-22180: Software to Assist in Referral of Pulmonary Fibrosis-Related Imaging Findings
Device definition: a prescription-use image processing software that analyzes computed tomography (CT) images to flag pulmonary fibrosis or imaging findings suggestive of the disease. Its output serves as adjunctive information in the referral pathway within the overall diagnostic assessment process.
| Control Item | FDA Special Control Requirement |
|---|---|
| Clinical performance | Premarket clinical performance validation and postmarket surveillance should demonstrate that the device performs as expected under the conditions of use for the intended patient population; FDA may determine whether postmarket surveillance data are needed based on the totality of premarket review information. |
| Reference standard validation | The report must describe in detail the reference standard, data, method used to define the standard, and selection rationale used to evaluate device performance, and demonstrate that the reference standard is appropriate for the disease indicated in the intended use. |
| Clinical comparator | Device performance must be compared against an appropriate clinical comparator, such as the performance or agreement of clinicians performing the same task. |
| Diagnostic metrics | Performance should be evaluated based on pre-specified diagnostic accuracy metrics; for example, ROC curves, sensitivity, specificity, positive/negative predictive values, and diagnostic likelihood ratios. |
| Patient subgroup testing | Data should include a sufficient number of cases from important subgroups to characterize performance estimates and confidence intervals for each subgroup, the target population, and imaging devices; this includes clinically relevant demographic characteristics, confounders, effect modifiers, comorbidities, early-stage disease, and image acquisition characteristics. |
| Software validation | Complete software verification, validation, and hazard analysis; software documentation should describe in detail the image analysis algorithm, model inputs and outputs, major component modules, and model limitations. |
| Labeling | Describe the clinical setting and workflow, compatible imaging hardware and acquisition protocols, methods and data set characteristics for each performance test, test results and confidence intervals, subgroup analyses, postmarket surveillance results, and device limitations, and advise users to interpret output in conjunction with other clinical and diagnostic findings. |
Key point: this document emphasizes clinical validation of the CT image analysis algorithm, appropriate clinical comparators, patient subgroup analyses, and continuous updating of labeling based on postmarket surveillance data.
2.3 Document 2026-22181: Spinal Imaging System for Neuraxial Anesthesia Procedures
Device definition: composed of sensors and software, used to assist in identifying intervertebral spaces during neuraxial anesthesia procedures.
| Control Item | FDA Special Control Requirement |
|---|---|
| Clinical performance | Demonstrate device performance under the conditions of use, and evaluate neuraxial procedure success rate, procedure time, and adverse events such as pain, bleeding, accidental dural puncture, and injury to other spinal structures. |
| Clinical comparator | Compare against manual palpation or other methods for identifying the target puncture site. |
| Nonclinical performance | Validate the accuracy, repeatability, and resolution of imaging of relevant anatomical structures representative of the intended use population. |
| Human factors | Demonstrate that users can use the device correctly based solely on the instructions for use. |
| Software | Complete software verification, validation, and hazard analysis; verification testing must confirm that images are constructed correctly. |
| Electrical and safety | Perform electromagnetic compatibility, electrical, mechanical, and thermal safety testing. |
| Biocompatibility | Demonstrate that patient-contacting components are biocompatible. |
| Sterility | For any device components labeled as sterile, performance testing must demonstrate sterility. |
| Labeling | Must warn that the risks inherent in neuraxial procedures under the standard of care remain even when the device is used. |
Key point: clinical validation focuses not only on image localization accuracy, but also requires evaluation of procedure success rate, procedure time, and adverse events; the device also cannot eliminate the risks inherent in neuraxial procedures.
2.4 Document 2026-22182: Over-the-Counter Sleep Apnea Risk Assessment Device
Device definition: uses software algorithms to analyze input sensor signals and flag the user’s risk of sleep apnea for users not previously diagnosed with sleep apnea. The device is not intended for stand-alone diagnosis, does not replace traditional diagnostic methods (such as polysomnography, PSG), is not intended to assist clinicians in diagnosing sleep disorders, and is not an apnea monitor.
| Control Item | FDA Special Control Requirement |
|---|---|
| Clinical performance | The study population should be representative of the intended use population, including symptomatic and asymptomatic subjects; any sample selection criteria or sample size limitations must be fully described and justified. |
| Data consistency | Evaluate the consistency of device output across the expected range of data sources and data quality in the intended use population and environment. |
| Patient subgroups | Evaluate device performance across all applicable subgroups, such as age, sex, BMI, and skin tone. |
| Clinical comparator | Compare against a clinical comparator device (such as PSG) to demonstrate that output metrics meet required accuracy and/or sensitivity and specificity; the basis for treating the clinical comparator as the reference standard must be described. |
| Machine learning algorithms | For devices using machine learning algorithms, clinical validation must use a database independent of the training data set. |
| Nonclinical performance | Validate hardware compatibility, the ability of software and hardware to provide adequate input signal quality, and the handling of noise, missing data, and low-quality signals. |
| Hardware safety | Provide electromagnetic compatibility and electrical, mechanical, and thermal safety performance data; skin-contacting hardware components must demonstrate biocompatibility. |
| Software documentation | Complete software verification, validation, and hazard analysis; describe software technical specifications, inputs and outputs of the detection algorithm, the influence of sensor acquisition hardware, and the impact of failures in subsystems for signal detection, analysis, display, and storage on output accuracy, as well as mitigation measures. |
| Human factors and usability | Demonstrate that users can use the device correctly based solely on the instructions for use, and can correctly interpret output and understand when to seek medical care. |
| Labeling | Describe what the device measures and outputs, hardware platform and operating system requirements, sensor data types, compatible hardware, factors affecting measurements, and limitations that the device cannot replace PSG or serve as an apnea monitor, and summarize overall and subgroup clinical performance. |
Key point: clinical validation data for machine learning algorithms must be independent of the training data set; the device must also help consumers correctly understand risk results and when to seek medical help.
3. What Are the Substantive Changes in the Four Documents?
From a regulatory perspective, the common changes in this update are mainly reflected in four areas: formal codification of the classification provisions into the CFR, a clear 510(k) pathway, more detailed special control requirements, and clearly defined boundaries of intended use.
3.1 Four New 21 CFR Classification Provisions
| Device Type | New Regulation | Classification |
|---|---|---|
| Skin cancer diagnostic assist software | 21 CFR 878.1830 | Class II |
| Pulmonary fibrosis imaging referral software | 21 CFR 892.2085 | Class II |
| Spinal imaging system | 21 CFR 868.1985 | Class II |
| Sleep apnea risk assessment device | 21 CFR 868.2378 | Class II |
Devices previously classified through the De Novo pathway now have formal CFR provisions, making device definitions and special control requirements more explicit.
3.2 All Four Device Types Are Subject to 510(k)
All four device types are Class II. FDA has not exempted these device types from 510(k) under Section 510(m) of the FD&C Act. Therefore, all four device types require a 510(k) premarket notification submission.
3.3 More Specific Special Control Requirements
| Focus Area | Requirements Stated in the Documents |
|---|---|
| Clinical validation | Explicit requirements for study population, reference standard, clinical comparator, evaluation metrics, and patient subgroup performance. |
| AI algorithms | Emphasis on algorithm inputs and outputs and model limitations; the sleep apnea device explicitly requires machine learning algorithms to use clinical validation data independent of the training set. |
| Human factors | Some devices explicitly require validation that users can use the device correctly and understand the output. |
| Software validation | Explicit requirements for software verification, validation, and hazard analysis. |
| Hardware compatibility | Explicit requirements to validate compatibility between the device and imaging or sensor acquisition hardware. |
| Labeling | Explicit requirements to disclose performance data, intended populations, limitations, risks of output interpretation and misuse. |
| Postmarket surveillance | The skin cancer diagnostic assist software and the pulmonary fibrosis imaging referral software explicitly involve postmarket surveillance requirements. |
3.4 Clear Boundaries of Intended Use
| Device | Boundary of Intended Use |
|---|---|
| Skin cancer diagnostic assist software | Must not be used as a stand-alone diagnostic device and must not be used to confirm a clinical diagnosis. |
| Pulmonary fibrosis imaging referral software | Output serves only as adjunctive information in the referral pathway within the overall diagnostic assessment process. |
| Spinal imaging system | Used to assist in identifying intervertebral spaces; using the device does not mean the risks inherent in neuraxial procedures disappear. |
| Sleep apnea risk assessment device | Cannot replace PSG, cannot be used for clinical diagnosis, and cannot serve as an apnea monitor. |
4. Side-by-Side Comparison of Special Control Requirements Across the Four Device Types
The following comparison is intended to quickly identify differences in special control items across the four device types. In the table, a dash (-) indicates that the document does not list it as a separate, explicit special control item; this does not mean that other general regulatory requirements do not apply.
| Test or Documentation Requirement | Skin Cancer Software | Pulmonary Fibrosis Software | Spinal Imaging System | Sleep Apnea Device |
|---|---|---|---|---|
| Clinical performance validation | Yes | Yes | Yes | Yes |
| Nonclinical performance testing | Yes | – | Yes | Yes |
| Software V&V and hazard analysis | Yes | Yes | Yes | Yes |
| Human factors/usability | Yes | – | Yes | Yes |
| Electrical safety/EMC | Yes | – | Yes | Yes |
| Biocompatibility | Yes | – | Yes | Yes |
| Sterility validation | – | – | Yes* | – |
| Cleaning and disinfection/reprocessing validation | Yes* | – | – | – |
| Postmarket surveillance | Yes | Yes | – | – |
Note: * indicates applicability only when corresponding conditions exist, such as reusable components or components labeled as sterile.
5. Implications for Medical Device Companies
In the context of FDA medical device registration work, the practical implications of this update for companies mainly include the following.
1. Regulatory requirements for AI medical devices are becoming more detailed. FDA looks not only at the algorithm itself, but also at training data, independent validation data, performance across patient subgroups, and the clinical significance of software output.
2. Clinical validation will be a major workstream. Different devices require appropriate clinical comparators, reference standards, and evaluation metrics; generic software performance testing cannot simply substitute for clinical performance validation.
3. Intended use directly affects the regulatory pathway and evidence requirements. There are clear distinctions among diagnostic assistance, risk flagging, clinical positioning, and stand-alone diagnosis; companies need to define the boundary of intended use at the product design and labeling stages.
4. 510(k) submissions require documentation tailored to the specific device type. If a device has the same or similar intended use and technical characteristics as the four device types above, companies may further investigate its feasibility as a potential predicate device or reference classification, but substantial equivalence requirements must still be met; these cannot simply be copied.
6. Conclusion
These four FDA final orders do not create a unified classification for all AI medical devices; rather, they formally establish Class II classifications and corresponding special control requirements for four specific device types.
Among them, the skin cancer diagnostic assist software focuses on clinical diagnostic accuracy and sensitivity for lesions with high metastatic potential; the pulmonary fibrosis software focuses on CT image algorithms, clinical validation, and postmarket surveillance; the spinal imaging system focuses on clinical procedure safety, imaging accuracy, and human factors; and the sleep apnea risk assessment device focuses on consumer use scenarios, sensor data quality, independent AI validation, and user understanding of risk.
For companies, the most important takeaway is not simply that all four device types are Class II, but to understand the specific clinical validation, software validation, performance testing, and labeling requirements that FDA has set for different AI and software-based medical devices.
Official Source Documents
- 2026-22179: Software to assist in the diagnosis of suspected skin cancer lesions
- 2026-22180: Software to assist in referral of pulmonary fibrosis-related imaging findings
- 2026-22181: Spinal imaging system for neuraxial anesthesia procedures
- 2026-22182: Over-the-counter sleep apnea risk assessment device
This article is compiled from FDA final classification orders and public Federal Register documents. For specific devices, classification, predicate device selection, and the applicability of special controls are subject to FDA’s current regulations and case-by-case review opinions.